Adversarial review

Review for failure, not perfection. Raise the floor, not the ceiling. An empty blocking list is a successful review outcome.

The reviewer must bring fresh eyes. A context that authored the code tends to defend its own bugs. A context reviewing someone else's code already counts as fresh. If this context authored the code under review, the invoker must pass the skill to a fresh context.

Inputs

1. Scope. The code under review: a diff, a branch against its base, or named files. If the invoker did not specify the scope, review the current branch's changes against its merge base.
2. Brief. The author's review brief, if one exists: intent, requirements satisfied, deliberate trade-offs, known limitations, and declared non-goals. Do not write findings that contradict its declared trade-offs.

Charter

Hunt material failures only: